Multi-cost based decision-theoretic rough sets in incomplete information systems

Xingbin Ma,Hengrong Ju,Xibei Yang,Jingjing Song
DOI: https://doi.org/10.13232/j.cnki.jnju.2015.02.018
2015-01-01
Abstract:Decision-theoretic rough set comes from Bayesian decision procedure,in which a pair of the thresholds is derived by the cost matrix for the construction of probabilistic rough set.Decision-theoretic rough set is a crucial rough set approach.By introducing the cost into probabilistic approximation of the target,the model of decision-theoretic rough set is actually sensitive to cost.However,most of the previous results about decision-theoretic rough set only use one and only one cost matrix to deal with the problems of the complete information systems.This method does not take the property of multiplicity and variability of cost into consideration,which is an important issue in machine learning and data mining.To solve such problems,a multi-cost strategy is firstly introduced into decision-theoretic rough set by using multiple cost matrixes.Moreover,the optimistic and pessimistic multi-cost decision-theoretic rough set models are proposed in incomplete information systems,respectively.Furthermore,the relationships are discussed between the two new decision-theoretic rough sets and the single cost matrix based decision-theoretic rough set.Then,we describe the formulas of the whole decision costs of optimistic and pessimistic multi-cost decision-theoretic rough set models.Finally,the several different decision costs of multi-cost decision-theoretic rough sets determined by decision-theoretic rough sets are tested on four UCI data sets.Experimental results show that the optimistic multi-cost decision-theoretic rough set model can generate the lowest decision cost. With the increase of the cost matrixes,all kinds of the whole decision costs will keep a steady value at last.The study suggests potential application areas and new research trends concerning decision-theoretic rough set.
What problem does this paper attempt to address?